# Business Central On-Premises Private AI Layer Blueprint Generator

> Source: https://netray.co/tools/business-central-on-prem-private-ai-blueprint-generator
> Publisher: Netray (netray.co), AI and integration engineering for ERP and manufacturing systems.
> Section: Free Tools

## Short answer

Describe your Business Central on-prem setup and get an architecture plan for a private AI assistant built on API v2.0 and a local model.

## Category

on-prem-ai

## Erp

Microsoft Dynamics 365 Business Central (on-premises)

## Audience

- CIO
- IT manager
- AL developer
- ERP architect

## Tool

**Kind:** template

**Output:** document

**Fields:**

#### Company name

**Key:** company

**Type:** text

**Default:** Ridgeway Fabrication Ltd

#### Business Central version

**Key:** version

**Type:** select

**Options:**

- Business Central 14
- 2019 release wave 2 (15)
- 2020 release wave 1 or 2 (16, 17)
- 2021 to 2023 releases (18 to 22)
- 2024 or later (23+)

**Default:** 2021 to 2023 releases (18 to 22)

#### Database engine

**Key:** db

**Type:** select

**Options:**

- SQL Server 2016
- SQL Server 2017
- SQL Server 2019
- SQL Server 2022

**Default:** SQL Server 2019

#### Hosting

**Key:** hosting

**Type:** select

**Options:**

- On-prem server
- Private cloud
- Air-gapped

**Default:** On-prem server

#### Named users

**Key:** users

**Type:** number

**Default:** 120

#### Companies or plants

**Key:** sites

**Type:** number

**Default:** 3

#### Main use cases

**Key:** usecases

**Type:** textarea

**Default:** Sales order and shipment status, item availability and reorder questions, vendor payment status, production order lookups, search of attached quotes and drawings.

#### Data that must never leave the site

**Key:** neverleave

**Type:** textarea

**Default:** Pricing, customer contracts, bank data, controlled technical data.

#### GPU available

**Key:** gpu

**Type:** select

**Options:**

- None yet
- One 24 GB class GPU
- One 48 to 80 GB class GPU
- Two or more GPUs

**Default:** One 48 to 80 GB class GPU

#### Target go-live

**Key:** golive

**Type:** text

**Default:** Q2 2027

**Blocks:**

#### Business Central Private AI Layer Plan: {{company}}

**Body:** System: {{version}} on {{db}}, hosting: {{hosting}}.
Scale: {{users}} named users, {{sites}} companies or plants.
Target go-live: {{golive}}.

#### Context

**Body:** {{usecases}}

Goal: a natural-language assistant over Business Central data and documents that runs wholly within {{company}} infrastructure.

#### Components

**Table:**

**Columns:**

- Component
- Design
- Notes

**Rows:**

**0:** Data access path

**1:** API v2.0 for standard entities; custom API pages in AL for gaps; OData web services where already published

**2:** Preferred over SQL for anything user-visible

**0:** Bulk reads

**1:** Read-only SQL replica for aggregates, validated against the application

**2:** Tables are per company and carry app ids

**0:** Retrieval index

**1:** Attachments, notes, external documents, with incremental refresh on modified dates

**2:** Local vector store

**0:** Model serving

**1:** Open-weight model via vLLM or Ollama, GPU: {{gpu}}

**2:** Plan for model updates

**0:** Tool server

**1:** MCP server with named read actions, parameter validation and row limits

**2:** No free-form query generation against production

**0:** Identity and permissions

**1:** Delegate user identity to API calls or enforce permission set filters in tool code

**2:** Test with low-privilege users

**0:** Audit

**1:** Local log of question, endpoint, parameters, answer, approver

**2:** Retain per policy

#### Security controls

**Table:**

**Columns:**

- Requirement
- Control

**Rows:**

**0:** Never leaves site: {{neverleave}}

**1:** No external APIs; inference, index and logs inside the {{hosting}} boundary; blocked egress

**0:** Transport

**1:** TLS on the server instance and tool server; certificates managed by IT

**0:** Permissions

**1:** Business Central permission sets honoured through delegated identity or equivalent filters

**0:** Write-back

**1:** Drafts only, human approval, POST through API

**0:** Traceability

**1:** Each answer cites the API call or view behind it

#### Three-phase rollout

**Table:**

**Columns:**

- Phase
- Scope
- Exit criteria

**Rows:**

**0:** 1. Read-only questions

**1:** Pilot group, standard APIs, document search

**2:** Agreed test set answered correctly; permissions verified

**0:** 2. Drafts

**1:** Draft quotes, purchase orders and notes for review

**2:** Low rework rate

**0:** 3. Approved actions

**1:** Few write actions through API with approval

**2:** Audit log reviewed by IT and finance

#### Open questions for IT

**Body:** 1. Which authentication method is configured on the server instance?
2. Which extensions add tables or fields users ask about?
3. Who can write and deploy custom API pages in AL?
4. Are API rate or resource limits a concern at peak?
5. Is a readable SQL replica available for aggregates?
6. Which Copilot or cloud features, if any, are available on this deployment and permitted?

**Signatures:**

- Prepared by
- IT architect
- Business owner
- Security reviewer

**Notes:** A starting plan. Check API availability, authentication options and extension behaviour for your Business Central build.

## Content

**Intro:** Business Central on-premises exposes a clean API surface, which makes a private AI layer easier than on older systems. This generator turns your environment details into a plan covering APIs, retrieval, model serving, permissions and rollout.

**Sections:**

#### How to use the generator

**Body:** Pick your Business Central release and database version, then list the use cases and the data that must stay on site. The output names the access path per question type, the tool server design, security controls and an open-questions list for your IT and AL developers.

**Bullets:**

- Know your authentication method
- List extensions with custom tables
- Decide delegation versus a service identity
- Export and review with security

#### The design behind the plan

**Body:** The plan uses API v2.0 and custom API pages for reads that must match what users see, a replica for aggregates, an index for documents and a local open-weight model. Permissions come from the calling identity, which is the main advantage of API-based access over SQL.

#### Common mistakes

**Body:** Using a broad service identity without per-user filtering, treating SQL columns as equal to API fields, pulling large sets through paged APIs and assuming cloud Copilot features exist on-premises. The plan lists each as a control or an open question.

**Cta:** If you would like help checking this plan against your extensions and server setup, Netray can review it with your team.

**Faq:**

#### Can I use this for Business Central online?

**Answer:** It is written for on-premises. An online tenant has different networking, authentication and data residency considerations.

#### Why prefer APIs over SQL?

**Answer:** APIs apply business logic and the caller's permissions. SQL reads raw tables, which are per company and may omit calculated values.

#### Does it need AL development?

**Answer:** Only if the standard API does not expose the data you need. Then a custom API page is the supported way to add it.

#### Is data stored by this tool?

**Answer:** No. The document is generated locally in your browser.

## Search terms

Business Central on-premises AI, Business Central private LLM architecture, Business Central API v2.0 AI plan, Business Central custom API page AI, on-prem Copilot alternative plan, air-gapped Business Central AI, Business Central permission sets AI

---
Talk to an engineer: https://netray.co/contact?intent=erp_ai | Machine catalog: https://netray.co/agent/catalog.json